fix: route Gemini image models through selector

docs/PROVIDERS.md

@@ -393,7 +393,7 @@ allowance). OpenMontage estimates cost from the transcribed audio duration. See
 
 ### Google — TTS + Imagen + Music + Video (Shared Key)
 
-> **One key, five tools.** Google Cloud TTS has 700+ voices in 50+ languages — the strongest localization option. Imagen 4 generates high-quality images. Google Lyria generates high-quality background music. Gemini Omni Flash supports conversational video editing, and direct Veo generation covers premium short video clips.
+> **One key, five tools.** Google Cloud TTS has 700+ voices in 50+ languages — the strongest localization option. `google_imagen` supports both Imagen 4 and Gemini 2.5 Flash Image, including projects without Imagen catalog access. Google Lyria generates high-quality background music. Gemini Omni Flash supports conversational video editing, and direct Veo generation covers premium short video clips.
 
 **Tools unlocked:** `google_tts`, `google_imagen`, `google_music`, `gemini_omni_video`, `veo_video`
 **Env var:** `GOOGLE_API_KEY` (or `GEMINI_API_KEY` — either works; `GEMINI_API_KEY` takes precedence)
@@ -434,9 +434,15 @@ The free tiers apply *independently* — you get 1M Standard AND 1M WaveNet AND
 | Imagen 4 Fast | $0.02 |
 | Imagen 4 Standard | $0.04 |
 | Imagen 4 Ultra | $0.06 |
+| Gemini 2.5 Flash Image (`gemini-2.5-flash-image`) | $0.039 |
 
 **Free tier for Imagen:** None. Paid tier only.
 
+To select the Gemini backend through the governed `image_selector`, pass
+`preferred_provider: "google_imagen"` and
+`model_name: "gemini-2.5-flash-image"`. The selector maps its neutral
+`model_name` field to the provider's `model` input.
+
 #### Gemini Omni Video Pricing
 
 | Model | Price | Notes |

tests/tools/test_google_imagen_gemini_backend.py

@@ -89,6 +89,31 @@ def test_gemini_model_routes_to_generate_content(imagen_tool, tmp_path):
     assert calls[0]["config"].image_config.aspect_ratio == "16:9"
 
 
+def test_image_selector_maps_model_name_to_google_model(
+    imagen_tool, monkeypatch, tmp_path
+):
+    """The governed selector must be able to reach the Gemini backend."""
+    from tools.graphics.image_selector import ImageSelector
+
+    tool, calls = imagen_tool
+    selector = ImageSelector()
+    monkeypatch.setattr(selector, "_providers", lambda: [tool])
+
+    result = selector.execute(
+        {
+            "prompt": "a flower",
+            "preferred_provider": "google_imagen",
+            "model_name": "gemini-2.5-flash-image",
+            "output_path": str(tmp_path / "selected.png"),
+        }
+    )
+
+    assert result.success, result.error
+    assert calls[0]["model"] == "gemini-2.5-flash-image"
+    assert result.data["selected_tool"] == "google_imagen"
+    assert result.data["model"] == "gemini-2.5-flash-image"
+
+
 def test_gemini_cost_estimate_is_per_image():
     from tools.graphics.google_imagen import GoogleImagen
 

tools/graphics/image_selector.py

@@ -242,6 +242,14 @@ class ImageSelector(BaseTool):
             props = tool.input_schema.get("properties", {})
             if "query" in props and "query" not in adapted:
                 adapted["query"] = adapted.get("prompt", "")
+            # The selector exposes a provider-neutral ``model_name`` field,
+            # while several providers call the same input ``model``.
+            if (
+                "model_name" in adapted
+                and "model" in props
+                and "model" not in adapted
+            ):
+                adapted["model"] = adapted["model_name"]
 
         # Strip selector-only keys that downstream tools don't understand
         adapted.pop("preferred_provider", None)